Session P36.6

Spectral Analysis of Atrial Signals Directly from Surface ECG Exploiting Compressed Spectrum

P Bonizzi*, O Meste, V Zarzoso

Université de Nice
Sophia Antipolis, France

Introduction: Atrial fibrillation (AF) represents the most common sustained cardiac arrhythmia in adults. Analysis of the AA extracted from surface ECG on subjects undergoing AF, in which the ventricular activity has first been cancelled, have been demonstrated to provide useful information on the characteristics of AF. In particular, a correlation between the spontaneous termination of the AF episode and the decreasing trend of the AF dominant frequency (AFDF) has been evidenced.
Methods: The characterization of the AFDF is facilitated if the AA signal is extracted from surface ECG. This extraction requires advanced signal processing techniques, characterized by a rather elevated computational cost. However, if evaluating the AFDF is the only purpose, this study shows that cancellation of the ventricular activity may be avoided by performing compressed spectral analysis directly on the surface ECG, without missing data problem. The basic principle of compressed spectrum (CS) is to accumulate the harmonics of the searched fundamental frequency, starting from an amplitude spectrum (AS) and a contraction of the frequency axis. The final result is that the amplitudes of the harmonics are added to the fundamental frequency, rather than to other frequencies, highlighting AFDF. A dataset composed of 22 recordings (all presenting AF) was employed to analyze the proposed idea. After appropriate pre-processing of ECG signal recorded in lead V1 (amongst others: removing baseline wander, truncation of QRST amplitude and band-pass filtering with passband 3-36Hz), CS was calculated in the 3-12Hz interval, till the third harmonic and AS was calculated for subsequent comparison. AFDFs estimated both by the proposed method and directly from AS were compared to those estimated on a reliable spatio-temporal QRST cancellation (ST-canc) method, taken as a reference.
Results and Conclusion: Differences in the AFDF estimates between the proposed method and the reference are negligible, as attested by their normalized mean square error (NMSE) of 1.0%. On the contrary, differences between AS-based AFDF estimate and the reference are larger, with a NMSE of 11.4% (AFDF, mean±SD (Hz): proposed method, 6.01±0.96; AS-based, 9.14±1.79;ST-canc, 6.19±1.02). We conclude that the AFDF estimation can be achieved without QRST cancellation. Moreover, the simplicity of the techniques makes it suitable for pseudo-real time applications.

(Abstract Control Number: 73)